[
https://issues.apache.org/jira/browse/CAMEL-24662?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Work on CAMEL-24662 started by Guillaume Nodet.
-----------------------------------------------
> camel-jbang: "camel kubernetes run --cluster-type X" is silently overridden
> by auto cluster detection
> -----------------------------------------------------------------------------------------------------
>
> Key: CAMEL-24662
> URL: https://issues.apache.org/jira/browse/CAMEL-24662
> Project: Camel
> Issue Type: Bug
> Components: camel-jbang
> Affects Versions: 4.22.0
> Reporter: Federico Mariani
> Assignee: Guillaume Nodet
> Priority: Major
>
> {{KubernetesRun.detectCluster()}} unconditionally overwrites the
> {{clusterType}} field with the result of auto-detection, discarding whatever
> the user explicitly passed via {{--cluster-type}}, unless {{--disable-auto}}
> is also given:
> {code:java}
> // dsl/camel-jbang/camel-jbang-plugin-kubernetes/.../KubernetesRun.java:809
> private void detectCluster() {
> if (!disableAuto) {
> ...
> ClusterType cluster = KubernetesHelper.discoverClusterType();
> this.clusterType = cluster.name(); // clobbers the explicit
> --cluster-type value
> if (ClusterType.MINIKUBE == cluster) {
> this.imageBuilder = "docker";
> this.imagePush = false;
> } else if (ClusterType.OPENSHIFT == cluster) {
> ...
> }
> }
> }
> {code}
> {{detectCluster()}} is called unconditionally from {{configureExport()}}
> before {{export.clusterType}} is read, so an explicit {{--cluster-type
> openshift}} (or any other value) only survives if auto-detection happens to
> agree with it. If the current kube context isn't reachable/detectable as
> OpenShift at CLI-invocation time (different network path than the actual
> deploy target, RBAC restricting the discovery call, CI without direct cluster
> access, etc.), the explicit flag is silently discarded and replaced with the
> auto-detected type - which also silently changes {{--image-builder}}
> (OpenShift forces {{docker}}, Minikube forces {{docker}} + disables push) and
> the jkube goal prefix ({{oc:}} vs {{k8s:}}) without any warning to the user.
> h3. Repro
> {noformat}
> camel kubernetes run --cluster-type openshift --output=yaml route.yaml
> application.properties
> {noformat}
> With a plain (non-OpenShift) kube context active, the generated project ends
> up using {{jkube.build.strategy=jib}} instead of {{docker}}, even though
> {{openshift}} was explicitly requested. Passing {{--disable-auto}} alongside
> {{--cluster-type openshift}} makes the explicit value stick (confirmed:
> {{jkube.build.strategy}} becomes {{docker}} and the deploy goal switches to
> {{oc:deploy}}).
> h3. Suggested fix
> Only run the auto-detection/override in {{detectCluster()}} when
> {{clusterType}} was not explicitly set by the user (e.g. track whether
> {{--cluster-type}} was passed, or seed the field with {{null}} and only
> assign when unset), instead of always overwriting it. At minimum, print a
> warning when the auto-detected type disagrees with an explicitly-provided
> {{--cluster-type}}.
> Found while investigating an unrelated report that
> {{camel.component.xslt-saxon.secureProcessing}} wasn't being honored on
> {{camel kubernetes run --cluster-type openshift}} deployments; that specific
> property-propagation behavior could not be reproduced (verified end-to-end
> across a plain Quarkus test, camel export, camel kubernetes run, a jib-built
> image, a live local Kubernetes deploy, and an OpenShift-style
> UBI/docker-built image - all correctly honored the property), so this ticket
> only covers the cluster-type override bug found along the way.
> _This issue was drafted by Claude Code on behalf of Federico Mariani_
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
